神经网络反向传播模型用于Cm_I的能级分类

Classification of Cm_Ⅰ Energy Levels Using the Back-propagation Neural Network

将神经网络反向传播算法应用于Ｃｕｒｉｕｍ的关于组态的分类计算。用４个特征变量（能级值、郎德ｇ日子、Ｊ量子数和同位素位移）表征每一个能好。为检验神经网络预报能力，采用ｌｅａｖｅ－ｎ－ｏｕｔ方法（奇宇称ｎ＝１，偶宇称ｎ＝５）对所有已知实验样本进行了预报。其中奇、偶宇称能级预报正确率分别为９２．７％和９７．４％。利用相同的神经网络结构，得到了１２个奇宇称和４２个偶宇称的未知能级的预报结果。

The back-propagation neural network was applied to the classification of the Cm_Ⅰ energy levels according to configuration. Four characteristic features,the energy level,Lande g factor,quantum number J,and isotope shift,were used to describe each level.To evaluate the predictive ability of the network,the leave-n-out method(n=1 for odd parity,and n=5 for even parity)was used to predict all of the levels.The predictive rates of odd and even levels are 92.7%and 97.4%,respectively. Using the same neural network architecture,prediction of containing 12-odd-parity and 42-even-partity unknow levels was obtained.